NIMR Tanzania MIP Analysis - Missigness

Overview

• 1,242 samples collected, captured, and sequenced from 13 locations across Tanzania
  • I have sequencing data for 1,237 of these. Notes on plate maps that some samples were missing.
• Genotyped using:
  • ~1800 MIP panel
    • Targets 1,800 SNPs across the genome (combination of neutral and under selection)
    • 3 total sequencing runs
  • Drug resistance MIP panel
    • Targets mutations in 13 genes associated with drug resistance
    • 2 total sequencing runs

Missingness - 1800 MIP Panel

• For 552 samples collected as part of the ‘Hotspots’ study, we have parasitemia (parasite density) data
• First plot below shows missingness (frequency of the 1800 target SNPs that failed to genotype), second plot shows the same thing after removing two outliers (>200k parasite density) stratified by location and plate:

• While parasitemia definetly contributes to high levels of missigness, some samples with high parasitemias are still failing.

• To see if expireiminatl failures contirbute to missingnes, I looked at missinges by plate (x axis) and target SNP position (y-axis):

• Plates 3 (Hotspots_P1), 4 (Hotspots_P2), and 12 (TES_2017_P8) all had high levels of missingess
• These plates contain samples from:
  • Chato (plates 3 & 4)
  • Nyanghwale and Uvinza (plates 4)
  • Ilemela and Kyela (plate 12)

• It clearly matters what plate a sample is on, and while parasitemia still contributes, it matters less if it wasn’t on one of plates 3, 4, or 12. For example, Uvinza samples, which were spread between plates 4, 5, and 6 had less missingness when they were on 5 compraed to when they were on 4:

• Here’s some simple linear regression models capturing this relationship.
• Here is missingness regressed on parasite-density and plate for the whole data set:

## 
## Call:
## lm(formula = hotspot2$F_MISS ~ hotspot2$pfdens + hotspot2$platenum)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -0.76159 -0.20166  0.03691  0.17336  0.77557 
## 
## Coefficients:
##                      Estimate Std. Error t value Pr(>|t|)    
## (Intercept)         9.517e-01  3.052e-02  31.178  < 2e-16 ***
## hotspot2$pfdens    -4.460e-06  6.939e-07  -6.427 2.86e-10 ***
## hotspot2$platenum4  2.345e-02  4.199e-02   0.558 0.576840    
## hotspot2$platenum5 -1.921e-01  4.212e-02  -4.561 6.31e-06 ***
## hotspot2$platenum6 -5.953e-01  4.216e-02 -14.120  < 2e-16 ***
## hotspot2$platenum7 -1.927e-01  4.202e-02  -4.587 5.59e-06 ***
## hotspot2$platenum8 -4.217e-01  4.198e-02 -10.045  < 2e-16 ***
## hotspot2$platenum9 -4.145e-01  1.193e-01  -3.474 0.000553 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.283 on 542 degrees of freedom
## Multiple R-squared:  0.4179, Adjusted R-squared:  0.4103 
## F-statistic: 55.58 on 7 and 542 DF,  p-value: < 2.2e-16

• Here’s the same model, but for just the UVINZA samples:

## 
## Call:
## lm(formula = uvinza$F_MISS ~ uvinza$pfdens + uvinza$platenum)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -0.68137 -0.19558  0.04626  0.13924  0.73034 
## 
## Coefficients:
##                    Estimate Std. Error t value Pr(>|t|)    
## (Intercept)       9.459e-01  4.911e-02  19.260  < 2e-16 ***
## uvinza$pfdens    -5.566e-06  1.405e-06  -3.960 0.000134 ***
## uvinza$platenum5 -1.797e-01  6.221e-02  -2.889 0.004666 ** 
## uvinza$platenum6 -5.900e-01  6.650e-02  -8.872 1.69e-14 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.274 on 108 degrees of freedom
## Multiple R-squared:  0.4695, Adjusted R-squared:  0.4548 
## F-statistic: 31.86 on 3 and 108 DF,  p-value: 7.84e-15

Missingness - Drug Resistance MIP Panel

• Here’s everything repeated, but for the drug resistance MIP panel.

Show missingness by parasitemia

Show missingness by plate

## 
## Call:
## lm(formula = hotspot2$F_MISS ~ hotspot2$pfdens)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -0.4035 -0.3508 -0.1759  0.4356  0.9586 
## 
## Coefficients:
##                   Estimate Std. Error t value Pr(>|t|)    
## (Intercept)      4.080e-01  1.936e-02  21.081  < 2e-16 ***
## hotspot2$pfdens -4.315e-06  9.530e-07  -4.528 7.31e-06 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.3931 on 548 degrees of freedom
## Multiple R-squared:  0.03607,    Adjusted R-squared:  0.03431 
## F-statistic:  20.5 on 1 and 548 DF,  p-value: 7.309e-06

## 
## Call:
## lm(formula = hotspot2$F_MISS ~ hotspot2$pfdens + hotspot2$platenum)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -0.67833 -0.23337 -0.07014  0.30740  0.89986 
## 
## Coefficients:
##                      Estimate Std. Error t value Pr(>|t|)    
## (Intercept)         2.622e-01  3.631e-02   7.219 1.77e-12 ***
## hotspot2$pfdens    -4.169e-06  8.255e-07  -5.050 6.06e-07 ***
## hotspot2$platenum4  4.267e-01  4.996e-02   8.542  < 2e-16 ***
## hotspot2$platenum5 -1.333e-01  5.011e-02  -2.661  0.00802 ** 
## hotspot2$platenum6  4.000e-01  5.015e-02   7.976 9.01e-15 ***
## hotspot2$platenum7  1.477e-01  4.999e-02   2.955  0.00327 ** 
## hotspot2$platenum8  3.719e-02  4.994e-02   0.745  0.45680    
## hotspot2$platenum9 -4.173e-02  1.419e-01  -0.294  0.76886    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.3367 on 542 degrees of freedom
## Multiple R-squared:  0.3005, Adjusted R-squared:  0.2914 
## F-statistic: 33.26 on 7 and 542 DF,  p-value: < 2.2e-16

Overall

• Here’s what we’re left with if we exclude samples with > 90% missing data for 1800 MIP panel:

• .. and for the DR MIP panel:

• Among the retained samples, some regions may still have higher amounts of missing genotypes tha nothers, which may be important to keep an eye on for certain popgen analyses with the 1800:

• Nyanghwale unfortunately was pretty unsalvagable… consider removing from manuscript?